阅读说明：本技术 一种基于自适应扩展卡尔曼滤波的电机参数估计方法 (Motor parameter estimation method based on adaptive extended Kalman filtering ) 是由 丁洁 曹正鑫 林金星 于 2020-06-23 设计创作，主要内容包括：本发明公开了一种自适应扩展卡尔曼滤波的电机参数估计方法,包括：将在αβ静止参考坐标系下的电流方程作为电机的连续状态空间表达式描述电机的内部动态特性；离散化状态空间方程同时将所需辨识参数增广到系统状态中；仿真电机采用双闭环控制方式,通过从电流检测单元和电压检测单元获得相电流、相电压；用当前时刻状态估计值代替系统真实值,并考虑满足白噪声时刻无关的性质,估计出当前时刻的系统噪声协方差矩阵；对噪声协方差估计值进行加权,得到当前时刻估计的噪声协方差矩阵；通过获得的相电流、相电压,利用已计算得到噪声协方差矩阵的扩展卡尔曼滤波算法估计反电动势同时辨识电机参数。该方法可提升电机参数估计的精度。(The invention discloses a motor parameter estimation method of adaptive extended Kalman filtering, which comprises the following steps: taking a current equation under an alpha beta static reference coordinate system as a continuous state space expression of the motor to describe the internal dynamic characteristics of the motor; the discretization state space equation simultaneously expands the needed identification parameters into the system state; the simulation motor adopts a double closed-loop control mode, and phase current and phase voltage are obtained from a current detection unit and a voltage detection unit; replacing a system true value with a current time state estimation value, and estimating a system noise covariance matrix at the current time by considering the property of meeting the white noise moment independence; weighting the noise covariance estimated value to obtain a noise covariance matrix estimated at the current moment; and estimating the back electromotive force and identifying the motor parameters by utilizing an extended Kalman filtering algorithm of the noise covariance matrix obtained by calculation through the obtained phase current and phase voltage. The method can improve the precision of motor parameter estimation.)

1. A motor parameter estimation method based on adaptive extended Kalman filtering is characterized in that:

the method comprises the following steps:

(1) taking a current equation under an alpha beta static reference coordinate system as a continuous state space expression of the motor to describe the internal dynamic characteristics of the motor;

(2) the discretization state space equation simultaneously expands the needed identification parameters into the system state;

(3) the simulation motor adopts a double closed-loop control mode, and phase current and phase voltage are obtained from a current detection unit and a voltage detection unit;

(4) replacing a system true value with a current time state estimation value, and estimating a system noise covariance matrix at the current time by considering the property of meeting the white noise moment independence; weighting the noise covariance estimated value to obtain a noise covariance matrix estimated at the current moment;

(5) and estimating the back electromotive force and identifying the motor parameters by utilizing an extended Kalman filtering algorithm of the noise covariance matrix obtained by calculation through the obtained phase current and phase voltage.

2. The motor parameter estimation method based on adaptive extended kalman filter according to claim 1, characterized in that: the current equation under the alpha beta static reference frame in the step (1) is as follows:

wherein i_α，i_βIs the stator current, e_α，e_βIs the back-emf of the winding, u_α，u_βIs the stator voltage in the αβ coordinate system;

r is stator resistance, L is stator inductance, which is equal to the difference between self inductance and mutual inductance of the winding,

the state space expression is expressed as:

wherein x₀＝[i_α，i_β，e_α，e_β]^TIs the system state vector, z ═ i_α，i_β]^TFor system output, u ═ u_α，u_β]^TIs the input of the system, and the system is,

3. the motor parameter estimation method based on adaptive extended kalman filter according to claim 2, characterized in that:

and (3) expanding L and R into a system state vector to obtain the following six-order model:

wherein x ═ x₀ ^T，L，R]^TAnd

obtaining a discrete time model by using a first-order Euler discretization method, and using T as the discrete time model_sIn order to be the sampling period of time,

A_k＝(I+T_sA)，B_k＝T_sB，H_k＝T_sH

since the upper and R are unknown, A_kx_k+B_ku_kIs a non-linear function of f_k(x_k，u_k) The non-linear function is represented by a linear function,

x_k＝f_k-1(x_k-1，u_k-1)+w_k-1

Z_k＝H_kx_k+v_k

wherein w_kAnd v_kSystem noise and measurement noise.

4. The motor parameter estimation method based on adaptive extended kalman filter according to claim 1, characterized in that: and (3) in a double closed loop control mode, the difference between the given value of the motor rotating speed and the actual rotating speed value calculated by the Hall signal is input as an input speed controller, the difference between the output of the speed controller and the current feedback quantity acquired by the current detection unit is input of the current controller, a PWM control signal generator judges the current motor rotor position according to the Hall sensor, and then the PWM control signal generator is connected to a power tube to be opened to finish the speed regulation of the motor, and meanwhile, phase current and phase voltage are obtained from the current detection unit and the voltage detection unit.

5. The motor parameter estimation method based on the adaptive extended kalman filter according to claim 1 or 2, wherein the specific process of the step (4) is as follows:

to estimate the noise covariance matrix Q_k-1X is to be_kBy using

further, it is possible to obtain:

at the same time, it is desirable to estimate process noise

finally, obtaining an estimation formula of the noise covariance:

to avoid certain system states, the corresponding coefficients in H are all 0, so thatCorresponding to the case where the value is 0,

where λ is a constant of (0, 1).

6. The motor parameter estimation method based on adaptive extended kalman filter according to claim 1, characterized in that: the estimation process is as follows:

K_k＝P_k/k-1H_k ^T(H_kP_k/k-1H_k ^T+R_k)^-1(3)

will obtain

P_k＝(I-K_kH_k)P_k/k-1。

Technical Field

The invention relates to a motor parameter estimation method, in particular to a motor parameter estimation method based on adaptive extended Kalman filtering.

Background

Brushless dc motor is widely used in various industrial equipments with its advantages such as long service life, simple control, reliable operation, etc., like: an industrial robot. At present, a plurality of controllers for controlling the brushless direct current motor are provided, such as various types of single chip microcomputers, DSPs, FPGAs and the like, and the controllers mostly adopt PWM and PID control modes. The inductance and resistance parameters are not only the precondition for realizing the precise control of the motor, but also the important basis for realizing the control algorithm of the frequency converter, analyzing the performance of the motor and optimizing the design, so in order to ensure the control precision and improve the performance of the motor, the motor parameters need to be identified in the actual engineering.

Brushless direct current motor (BLDCM) with its advantage such as long-lived, control is simple, operation are reliable, the wide application is in industrial equipment such as industrial robot, digit control machine tool. At present, the hot spots of the brushless dc motor are mainly the design and control thereof, including the design of the brushless motor body, torque ripple, sensorless control, torque control, and the like. The key of most realizing sensorless control and torque control is to obtain accurate and real-time back electromotive force of the motor, generally, the back electromotive force of the motor is considered to be ideal trapezoidal wave, but the control precision is low; or calculating the back emf value in the control scheme by table look-up, however, this adds an operational step.

When the motor back emf is taken as the state variable, the estimation can be done with a state observer or filter. The problem is that the state estimation method relies on accurate motor parameters. For a brushless direct current motor, the current change is large in the commutation process, and the deviation of inductance and resistance parameters has great influence on back electromotive force estimation. Therefore, in order to ensure the control precision and improve the motor performance, it is necessary to identify the motor parameters in engineering practice. The parameter identification method mainly comprises a least square method, sliding mode identification, a model reference self-adaption method, an artificial neural network, a Kalman filter and the like. Kalman filtering is an effective filtering method for multidimensional states and non-stationary processes, with high accuracy even in the presence of noisy disturbances. Due to the non-linearity, it is one of the most common methods in motor parameter identification. The extended Kalman filtering algorithm is simple to calculate and easy to track. However, since only the first order term of the Taylor expansion is retained, the result obtained by the algorithm may have errors, and the stronger the nonlinearity degree of the system, the larger the error of the predicted result is. Unknown noise or poor estimation accuracy may affect the accuracy control of the motor.

Disclosure of Invention

The invention aims to provide a motor parameter estimation method based on adaptive extended Kalman filtering, which has high accuracy and does not generate larger deviation and divergence during filtering.

The invention discloses a motor parameter estimation method based on adaptive extended Kalman filtering, which comprises the following steps:

(1) taking a current equation under an alpha beta static reference coordinate system as a continuous state space expression of the motor to describe the internal dynamic characteristics of the motor;

(2) the discretization state space equation simultaneously expands the needed identification parameters into the system state;

(3) the simulation motor adopts a double closed-loop control mode, and phase current and phase voltage are obtained from a current detection unit and a voltage detection unit;

(4) replacing a system true value with a current time state estimation value, and estimating a system noise covariance matrix at the current time by considering the property of meeting the white noise moment independence; weighting the noise covariance estimated value to obtain a noise covariance matrix estimated at the current moment;

(5) and estimating the back electromotive force and identifying the motor parameters by utilizing an extended Kalman filtering algorithm of the noise covariance matrix obtained by calculation through the obtained phase current and phase voltage.

Further, the current equation in the α β stationary reference frame in step (1) is as follows:

wherein i_α，i_βIs the stator current, e_α，e_βIs the back-emf of the winding, u_α，u_βIs the stator voltage in the αβ coordinate system;

r is stator resistance, L is stator inductance, which is equal to the difference between self inductance and mutual inductance of the winding,

the state space expression is expressed as:

wherein x₀＝[i_α，i_α，e_α，e_β]^TIs the system state vector, z ═ i_α，i_β]^TFor system output, u ═ u_α，u_β]^TIs input to the system, and

further, L and R are augmented into the system state vector, resulting in the following sixth order model:

wherein x ═ x₀ ^T，L，R]^TAnd

obtaining a discrete time model by using a first-order Euler discretization method, and using T as the discrete time model_sIn order to be the sampling period of time,

A_k＝(I+T_sA)，B_k＝T_sB，H_k＝T_sH

since L and R are unknown, A_kx_k+B_ku_kIs a non-linear function of f_k(x_k，u_k) The non-linear function is represented by a linear function,

x_k＝f_k-1(x_k-1，u_k-1)+w_k-1

Z_k＝H_kx_k+v_k

wherein w_kAnd v_kSystem noise and measurement noise.

Further, the simulation motor in the step (3) is controlled by adopting a two-by-two 120-degree conduction mode, in a double closed loop control mode, the difference between the given value of the motor rotating speed and the actual rotating speed value calculated by the Hall signal is input into the speed controller, the difference between the output of the speed controller and the current feedback quantity acquired by the current detection unit is input into the current controller, the PWM control signal generator judges the current motor rotor position according to the Hall sensor, and then the PWM control signal generator is connected to a power tube to be opened to finish the speed regulation of the motor, and meanwhile, the phase current and the phase voltage are obtained from the current detection unit and the voltage detection unit.

Further, the specific process of the step (4) is as follows:

to estimate the noise covariance matrix Q_k-1X is to be_kBy usingInstead, one can obtain:

further, it is possible to obtain:

at the same time, it is desirable to estimate process noiseThe different moments are independent, namely:

finally, obtaining an estimation formula of the noise covariance:

to avoid certain system states, the corresponding coefficients in H are all 0, so thatCorresponding to the case where the value is 0,

where λ is a constant of (0, 1).

Further, the estimation process is as follows:

K_k＝P_k/k-1H_k ^T(H_kP_k/k-1H_k ^T+R_k)^-1(3)

will obtain

Replace the original

Back to the formulas (2) - (4)

P_k＝(I-K_kH_k)P_k/k-1。

Has the advantages that: the method fully utilizes the input and output data of the motor model, improves the precision of parameter estimation, is simple and easy to realize, has small filtering deviation and is not easy to disperse.

Drawings

Fig. 1 is a block flow diagram illustrating a method for estimating parameters of a brushless dc motor based on adaptive extended kalman filter according to an embodiment of the present invention;

FIG. 2 is a control frame diagram of a brushless DC motor in simulation according to the present invention;

FIG. 3 shows the EKF and AEKF vs. the motor status e in the simulation of the present invention_αA comparison graph of the estimated values;

FIG. 4 is a graph of estimated values and errors of EKF and AEKF on motor parameters L in simulation of the present invention;

FIG. 5 is a graph of estimated values and errors of EKF and AEKF on motor parameters R in simulation of the present invention.

Detailed Description

The parameter estimation method of the brushless direct current motor based on the adaptive extended kalman filter is provided by the embodiment.

Referring to fig. 1, the method specifically includes the following steps:

1. and describing the internal dynamic characteristics of the motor by using a current equation under an alpha beta static reference frame as a continuous state space expression of the motor.

2. The discretization state space equation simultaneously expands the needed identification parameters into the system state; firstly, taking motor parameters L and R as motor states to be expanded to the original motor states to obtain a six-order model:wherein x ═ x₀ ^T，L，R]^TAnd

H＝[H₀，0_2×2](ii) a Secondly, a discrete time model is obtained by utilizing first-order Euler discretization and is represented by T_sIs a sampling period: a. the_k＝(I+T_sA)，B_k＝T_sB，H_k＝T_sH. Since L and R are unknown, A_kx_k+B_ku_kIs a non-linear function. By f_k(x_k，u_k) Represents a non-linear function: x is the number of_k＝f_k-1(x_k-1，u_k-1)+w_k-1，z_k＝H_kx_k+v_k

Wherein w_kAnd v_kSystem noise and measurement noise.

3. The experimental motor of the permanent magnet synchronous direct current motor in the simulation system is a surface-mounted permanent magnet brushless direct current motor, the two-two 120-degree conduction mode control is adopted, a double-closed-loop control mode is used, and as shown in fig. 2, the difference between the given value of the motor rotating speed and the actual rotating speed value calculated by the Hall signal is used as the input of an input speed controller. The difference between the output of the speed controller and the current feedback quantity acquired by the current detection unit is the input of the current controller. And the PWM control signal generator can interpret the current position of the motor rotor according to the Hall sensor and then is connected to a power tube to be opened so as to finish the speed regulation of the motor. Meanwhile, the phase current and the phase voltage are obtained by obtaining them from the current detection unit and the voltage detection unit.

4. Known from the traditional extended kalman filter algorithm:

to estimate Q_k-1X is to be_kBy using

Instead, one can obtain:

further, it is possible to obtain:

at the same time, it is desirable to estimate process noise

The different moments are independent, namely:

is equivalent to: (I-K)_kH_k)F_k-1P_k-1F_k-1 ^T(H_k ^TK_k ^T-I)+(I-K_kH_k)Q_k-1H_k ^TK_k ^T-K_kR_kK_k ^TAvailable as 0:use ofIn place of Q_k-1The modified noise covariance may be modified to

To avoid certain system states, the corresponding coefficients in H are all 0, so that

Case where the corresponding value is 0:wherein λ is a constant of (0, 1)

5. Setting the sampling period T_s＝2×10^-6λ is 0.7, and each initial value is x₀＝[0，0，0，0，0.01，0.5]^T，P₀＝diag[1，1，1，1，1，1]^T，Q₀Biag (0.01, 0.01, 10, 10, 0, 0) R ═ diag (1000 ). And respectively using an extended Kalman filtering algorithm and a self-adaptive extended Kalman filtering algorithm to estimate the back electromotive force of the motor and simultaneously identify motor parameters through the obtained phase current and phase voltage. As can be seen from fig. 3, the error of the back electromotive force estimated by the two algorithms is gradually reduced, but the initial error estimated by the extended kalman filter algorithm is relatively large, and the error estimated by the adaptive extended kalman filter algorithm is slightly smaller than that of the former; as can be seen from fig. 4, the errors of the motor parameters L estimated by the two algorithms are both small, and the convergence speed is high, but the error estimated by the adaptive extended kalman filter algorithm is smaller; as can be seen from FIG. 5, the R precision estimated by using the adaptive extended Kalman filtering algorithm is greatly improved compared with the traditional extended Kalman filtering method, and the convergence rate is higher. Compared with the prior art, the method has better accuracy in estimating the parameters of the brushless direct current motor, thereby improving the accuracy of the estimation of the back electromotive force of the motor and improving the control accuracy of the motor.